xref: /openbmc/qemu/hw/i386/x86-common.c (revision 5304873a)
1 /*
2  * Copyright (c) 2003-2004 Fabrice Bellard
3  * Copyright (c) 2019, 2024 Red Hat, Inc.
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a copy
6  * of this software and associated documentation files (the "Software"), to deal
7  * in the Software without restriction, including without limitation the rights
8  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9  * copies of the Software, and to permit persons to whom the Software is
10  * furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice shall be included in
13  * all copies or substantial portions of the Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
21  * THE SOFTWARE.
22  */
23 #include "qemu/osdep.h"
24 #include "qemu/error-report.h"
25 #include "qemu/cutils.h"
26 #include "qemu/units.h"
27 #include "qemu/datadir.h"
28 #include "qapi/error.h"
29 #include "sysemu/numa.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/xen.h"
32 #include "trace.h"
33 
34 #include "hw/i386/x86.h"
35 #include "target/i386/cpu.h"
36 #include "hw/rtc/mc146818rtc.h"
37 #include "target/i386/sev.h"
38 
39 #include "hw/acpi/cpu_hotplug.h"
40 #include "hw/irq.h"
41 #include "hw/loader.h"
42 #include "multiboot.h"
43 #include "elf.h"
44 #include "standard-headers/asm-x86/bootparam.h"
45 #include CONFIG_DEVICES
46 #include "kvm/kvm_i386.h"
47 
48 #ifdef CONFIG_XEN_EMU
49 #include "hw/xen/xen.h"
50 #include "hw/i386/kvm/xen_evtchn.h"
51 #endif
52 
53 /* Physical Address of PVH entry point read from kernel ELF NOTE */
54 static size_t pvh_start_addr;
55 
56 static void x86_cpu_new(X86MachineState *x86ms, int64_t apic_id, Error **errp)
57 {
58     Object *cpu = object_new(MACHINE(x86ms)->cpu_type);
59 
60     if (!object_property_set_uint(cpu, "apic-id", apic_id, errp)) {
61         goto out;
62     }
63     qdev_realize(DEVICE(cpu), NULL, errp);
64 
65 out:
66     object_unref(cpu);
67 }
68 
69 void x86_cpus_init(X86MachineState *x86ms, int default_cpu_version)
70 {
71     int i;
72     const CPUArchIdList *possible_cpus;
73     MachineState *ms = MACHINE(x86ms);
74     MachineClass *mc = MACHINE_GET_CLASS(x86ms);
75 
76     x86_cpu_set_default_version(default_cpu_version);
77 
78     /*
79      * Calculates the limit to CPU APIC ID values
80      *
81      * Limit for the APIC ID value, so that all
82      * CPU APIC IDs are < x86ms->apic_id_limit.
83      *
84      * This is used for FW_CFG_MAX_CPUS. See comments on fw_cfg_arch_create().
85      */
86     x86ms->apic_id_limit = x86_cpu_apic_id_from_index(x86ms,
87                                                       ms->smp.max_cpus - 1) + 1;
88 
89     /*
90      * Can we support APIC ID 255 or higher?  With KVM, that requires
91      * both in-kernel lapic and X2APIC userspace API.
92      *
93      * kvm_enabled() must go first to ensure that kvm_* references are
94      * not emitted for the linker to consume (kvm_enabled() is
95      * a literal `0` in configurations where kvm_* aren't defined)
96      */
97     if (kvm_enabled() && x86ms->apic_id_limit > 255 &&
98         kvm_irqchip_in_kernel() && !kvm_enable_x2apic()) {
99         error_report("current -smp configuration requires kernel "
100                      "irqchip and X2APIC API support.");
101         exit(EXIT_FAILURE);
102     }
103 
104     if (kvm_enabled()) {
105         kvm_set_max_apic_id(x86ms->apic_id_limit);
106     }
107 
108     if (!kvm_irqchip_in_kernel()) {
109         apic_set_max_apic_id(x86ms->apic_id_limit);
110     }
111 
112     possible_cpus = mc->possible_cpu_arch_ids(ms);
113     for (i = 0; i < ms->smp.cpus; i++) {
114         x86_cpu_new(x86ms, possible_cpus->cpus[i].arch_id, &error_fatal);
115     }
116 }
117 
118 void x86_rtc_set_cpus_count(ISADevice *s, uint16_t cpus_count)
119 {
120     MC146818RtcState *rtc = MC146818_RTC(s);
121 
122     if (cpus_count > 0xff) {
123         /*
124          * If the number of CPUs can't be represented in 8 bits, the
125          * BIOS must use "FW_CFG_NB_CPUS". Set RTC field to 0 just
126          * to make old BIOSes fail more predictably.
127          */
128         mc146818rtc_set_cmos_data(rtc, 0x5f, 0);
129     } else {
130         mc146818rtc_set_cmos_data(rtc, 0x5f, cpus_count - 1);
131     }
132 }
133 
134 static int x86_apic_cmp(const void *a, const void *b)
135 {
136    CPUArchId *apic_a = (CPUArchId *)a;
137    CPUArchId *apic_b = (CPUArchId *)b;
138 
139    return apic_a->arch_id - apic_b->arch_id;
140 }
141 
142 /*
143  * returns pointer to CPUArchId descriptor that matches CPU's apic_id
144  * in ms->possible_cpus->cpus, if ms->possible_cpus->cpus has no
145  * entry corresponding to CPU's apic_id returns NULL.
146  */
147 static CPUArchId *x86_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
148 {
149     CPUArchId apic_id, *found_cpu;
150 
151     apic_id.arch_id = id;
152     found_cpu = bsearch(&apic_id, ms->possible_cpus->cpus,
153         ms->possible_cpus->len, sizeof(*ms->possible_cpus->cpus),
154         x86_apic_cmp);
155     if (found_cpu && idx) {
156         *idx = found_cpu - ms->possible_cpus->cpus;
157     }
158     return found_cpu;
159 }
160 
161 void x86_cpu_plug(HotplugHandler *hotplug_dev,
162                   DeviceState *dev, Error **errp)
163 {
164     CPUArchId *found_cpu;
165     Error *local_err = NULL;
166     X86CPU *cpu = X86_CPU(dev);
167     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
168 
169     if (x86ms->acpi_dev) {
170         hotplug_handler_plug(x86ms->acpi_dev, dev, &local_err);
171         if (local_err) {
172             goto out;
173         }
174     }
175 
176     /* increment the number of CPUs */
177     x86ms->boot_cpus++;
178     if (x86ms->rtc) {
179         x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
180     }
181     if (x86ms->fw_cfg) {
182         fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
183     }
184 
185     found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
186     found_cpu->cpu = CPU(dev);
187 out:
188     error_propagate(errp, local_err);
189 }
190 
191 void x86_cpu_unplug_request_cb(HotplugHandler *hotplug_dev,
192                                DeviceState *dev, Error **errp)
193 {
194     int idx = -1;
195     X86CPU *cpu = X86_CPU(dev);
196     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
197 
198     if (!x86ms->acpi_dev) {
199         error_setg(errp, "CPU hot unplug not supported without ACPI");
200         return;
201     }
202 
203     x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
204     assert(idx != -1);
205     if (idx == 0) {
206         error_setg(errp, "Boot CPU is unpluggable");
207         return;
208     }
209 
210     hotplug_handler_unplug_request(x86ms->acpi_dev, dev,
211                                    errp);
212 }
213 
214 void x86_cpu_unplug_cb(HotplugHandler *hotplug_dev,
215                        DeviceState *dev, Error **errp)
216 {
217     CPUArchId *found_cpu;
218     Error *local_err = NULL;
219     X86CPU *cpu = X86_CPU(dev);
220     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
221 
222     hotplug_handler_unplug(x86ms->acpi_dev, dev, &local_err);
223     if (local_err) {
224         goto out;
225     }
226 
227     found_cpu = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, NULL);
228     found_cpu->cpu = NULL;
229     qdev_unrealize(dev);
230 
231     /* decrement the number of CPUs */
232     x86ms->boot_cpus--;
233     /* Update the number of CPUs in CMOS */
234     x86_rtc_set_cpus_count(x86ms->rtc, x86ms->boot_cpus);
235     fw_cfg_modify_i16(x86ms->fw_cfg, FW_CFG_NB_CPUS, x86ms->boot_cpus);
236  out:
237     error_propagate(errp, local_err);
238 }
239 
240 void x86_cpu_pre_plug(HotplugHandler *hotplug_dev,
241                       DeviceState *dev, Error **errp)
242 {
243     int idx;
244     CPUState *cs;
245     CPUArchId *cpu_slot;
246     X86CPUTopoIDs topo_ids;
247     X86CPU *cpu = X86_CPU(dev);
248     CPUX86State *env = &cpu->env;
249     MachineState *ms = MACHINE(hotplug_dev);
250     X86MachineState *x86ms = X86_MACHINE(hotplug_dev);
251     unsigned int smp_cores = ms->smp.cores;
252     unsigned int smp_threads = ms->smp.threads;
253     X86CPUTopoInfo topo_info;
254 
255     if (!object_dynamic_cast(OBJECT(cpu), ms->cpu_type)) {
256         error_setg(errp, "Invalid CPU type, expected cpu type: '%s'",
257                    ms->cpu_type);
258         return;
259     }
260 
261     if (x86ms->acpi_dev) {
262         Error *local_err = NULL;
263 
264         hotplug_handler_pre_plug(HOTPLUG_HANDLER(x86ms->acpi_dev), dev,
265                                  &local_err);
266         if (local_err) {
267             error_propagate(errp, local_err);
268             return;
269         }
270     }
271 
272     init_topo_info(&topo_info, x86ms);
273 
274     if (ms->smp.modules > 1) {
275         env->nr_modules = ms->smp.modules;
276         set_bit(CPU_TOPO_LEVEL_MODULE, env->avail_cpu_topo);
277     }
278 
279     if (ms->smp.dies > 1) {
280         env->nr_dies = ms->smp.dies;
281         set_bit(CPU_TOPO_LEVEL_DIE, env->avail_cpu_topo);
282     }
283 
284     /*
285      * If APIC ID is not set,
286      * set it based on socket/die/core/thread properties.
287      */
288     if (cpu->apic_id == UNASSIGNED_APIC_ID) {
289         int max_socket = (ms->smp.max_cpus - 1) /
290                                 smp_threads / smp_cores / ms->smp.dies;
291 
292         /*
293          * die-id was optional in QEMU 4.0 and older, so keep it optional
294          * if there's only one die per socket.
295          */
296         if (cpu->die_id < 0 && ms->smp.dies == 1) {
297             cpu->die_id = 0;
298         }
299 
300         if (cpu->socket_id < 0) {
301             error_setg(errp, "CPU socket-id is not set");
302             return;
303         } else if (cpu->socket_id > max_socket) {
304             error_setg(errp, "Invalid CPU socket-id: %u must be in range 0:%u",
305                        cpu->socket_id, max_socket);
306             return;
307         }
308         if (cpu->die_id < 0) {
309             error_setg(errp, "CPU die-id is not set");
310             return;
311         } else if (cpu->die_id > ms->smp.dies - 1) {
312             error_setg(errp, "Invalid CPU die-id: %u must be in range 0:%u",
313                        cpu->die_id, ms->smp.dies - 1);
314             return;
315         }
316         if (cpu->core_id < 0) {
317             error_setg(errp, "CPU core-id is not set");
318             return;
319         } else if (cpu->core_id > (smp_cores - 1)) {
320             error_setg(errp, "Invalid CPU core-id: %u must be in range 0:%u",
321                        cpu->core_id, smp_cores - 1);
322             return;
323         }
324         if (cpu->thread_id < 0) {
325             error_setg(errp, "CPU thread-id is not set");
326             return;
327         } else if (cpu->thread_id > (smp_threads - 1)) {
328             error_setg(errp, "Invalid CPU thread-id: %u must be in range 0:%u",
329                        cpu->thread_id, smp_threads - 1);
330             return;
331         }
332 
333         topo_ids.pkg_id = cpu->socket_id;
334         topo_ids.die_id = cpu->die_id;
335         topo_ids.core_id = cpu->core_id;
336         topo_ids.smt_id = cpu->thread_id;
337         cpu->apic_id = x86_apicid_from_topo_ids(&topo_info, &topo_ids);
338     }
339 
340     cpu_slot = x86_find_cpu_slot(MACHINE(x86ms), cpu->apic_id, &idx);
341     if (!cpu_slot) {
342         x86_topo_ids_from_apicid(cpu->apic_id, &topo_info, &topo_ids);
343         error_setg(errp,
344             "Invalid CPU [socket: %u, die: %u, core: %u, thread: %u] with"
345             " APIC ID %" PRIu32 ", valid index range 0:%d",
346             topo_ids.pkg_id, topo_ids.die_id, topo_ids.core_id, topo_ids.smt_id,
347             cpu->apic_id, ms->possible_cpus->len - 1);
348         return;
349     }
350 
351     if (cpu_slot->cpu) {
352         error_setg(errp, "CPU[%d] with APIC ID %" PRIu32 " exists",
353                    idx, cpu->apic_id);
354         return;
355     }
356 
357     /* if 'address' properties socket-id/core-id/thread-id are not set, set them
358      * so that machine_query_hotpluggable_cpus would show correct values
359      */
360     /* TODO: move socket_id/core_id/thread_id checks into x86_cpu_realizefn()
361      * once -smp refactoring is complete and there will be CPU private
362      * CPUState::nr_cores and CPUState::nr_threads fields instead of globals */
363     x86_topo_ids_from_apicid(cpu->apic_id, &topo_info, &topo_ids);
364     if (cpu->socket_id != -1 && cpu->socket_id != topo_ids.pkg_id) {
365         error_setg(errp, "property socket-id: %u doesn't match set apic-id:"
366             " 0x%x (socket-id: %u)", cpu->socket_id, cpu->apic_id,
367             topo_ids.pkg_id);
368         return;
369     }
370     cpu->socket_id = topo_ids.pkg_id;
371 
372     if (cpu->die_id != -1 && cpu->die_id != topo_ids.die_id) {
373         error_setg(errp, "property die-id: %u doesn't match set apic-id:"
374             " 0x%x (die-id: %u)", cpu->die_id, cpu->apic_id, topo_ids.die_id);
375         return;
376     }
377     cpu->die_id = topo_ids.die_id;
378 
379     if (cpu->core_id != -1 && cpu->core_id != topo_ids.core_id) {
380         error_setg(errp, "property core-id: %u doesn't match set apic-id:"
381             " 0x%x (core-id: %u)", cpu->core_id, cpu->apic_id,
382             topo_ids.core_id);
383         return;
384     }
385     cpu->core_id = topo_ids.core_id;
386 
387     if (cpu->thread_id != -1 && cpu->thread_id != topo_ids.smt_id) {
388         error_setg(errp, "property thread-id: %u doesn't match set apic-id:"
389             " 0x%x (thread-id: %u)", cpu->thread_id, cpu->apic_id,
390             topo_ids.smt_id);
391         return;
392     }
393     cpu->thread_id = topo_ids.smt_id;
394 
395     /*
396     * kvm_enabled() must go first to ensure that kvm_* references are
397     * not emitted for the linker to consume (kvm_enabled() is
398     * a literal `0` in configurations where kvm_* aren't defined)
399     */
400     if (kvm_enabled() && hyperv_feat_enabled(cpu, HYPERV_FEAT_VPINDEX) &&
401         !kvm_hv_vpindex_settable()) {
402         error_setg(errp, "kernel doesn't allow setting HyperV VP_INDEX");
403         return;
404     }
405 
406     cs = CPU(cpu);
407     cs->cpu_index = idx;
408 
409     numa_cpu_pre_plug(cpu_slot, dev, errp);
410 }
411 
412 static long get_file_size(FILE *f)
413 {
414     long where, size;
415 
416     /* XXX: on Unix systems, using fstat() probably makes more sense */
417 
418     where = ftell(f);
419     fseek(f, 0, SEEK_END);
420     size = ftell(f);
421     fseek(f, where, SEEK_SET);
422 
423     return size;
424 }
425 
426 void gsi_handler(void *opaque, int n, int level)
427 {
428     GSIState *s = opaque;
429 
430     trace_x86_gsi_interrupt(n, level);
431     switch (n) {
432     case 0 ... ISA_NUM_IRQS - 1:
433         if (s->i8259_irq[n]) {
434             /* Under KVM, Kernel will forward to both PIC and IOAPIC */
435             qemu_set_irq(s->i8259_irq[n], level);
436         }
437         /* fall through */
438     case ISA_NUM_IRQS ... IOAPIC_NUM_PINS - 1:
439 #ifdef CONFIG_XEN_EMU
440         /*
441          * Xen delivers the GSI to the Legacy PIC (not that Legacy PIC
442          * routing actually works properly under Xen). And then to
443          * *either* the PIRQ handling or the I/OAPIC depending on
444          * whether the former wants it.
445          */
446         if (xen_mode == XEN_EMULATE && xen_evtchn_set_gsi(n, level)) {
447             break;
448         }
449 #endif
450         qemu_set_irq(s->ioapic_irq[n], level);
451         break;
452     case IO_APIC_SECONDARY_IRQBASE
453         ... IO_APIC_SECONDARY_IRQBASE + IOAPIC_NUM_PINS - 1:
454         qemu_set_irq(s->ioapic2_irq[n - IO_APIC_SECONDARY_IRQBASE], level);
455         break;
456     }
457 }
458 
459 void ioapic_init_gsi(GSIState *gsi_state, Object *parent)
460 {
461     DeviceState *dev;
462     SysBusDevice *d;
463     unsigned int i;
464 
465     assert(parent);
466     if (kvm_ioapic_in_kernel()) {
467         dev = qdev_new(TYPE_KVM_IOAPIC);
468     } else {
469         dev = qdev_new(TYPE_IOAPIC);
470     }
471     object_property_add_child(parent, "ioapic", OBJECT(dev));
472     d = SYS_BUS_DEVICE(dev);
473     sysbus_realize_and_unref(d, &error_fatal);
474     sysbus_mmio_map(d, 0, IO_APIC_DEFAULT_ADDRESS);
475 
476     for (i = 0; i < IOAPIC_NUM_PINS; i++) {
477         gsi_state->ioapic_irq[i] = qdev_get_gpio_in(dev, i);
478     }
479 }
480 
481 DeviceState *ioapic_init_secondary(GSIState *gsi_state)
482 {
483     DeviceState *dev;
484     SysBusDevice *d;
485     unsigned int i;
486 
487     dev = qdev_new(TYPE_IOAPIC);
488     d = SYS_BUS_DEVICE(dev);
489     sysbus_realize_and_unref(d, &error_fatal);
490     sysbus_mmio_map(d, 0, IO_APIC_SECONDARY_ADDRESS);
491 
492     for (i = 0; i < IOAPIC_NUM_PINS; i++) {
493         gsi_state->ioapic2_irq[i] = qdev_get_gpio_in(dev, i);
494     }
495     return dev;
496 }
497 
498 /*
499  * The entry point into the kernel for PVH boot is different from
500  * the native entry point.  The PVH entry is defined by the x86/HVM
501  * direct boot ABI and is available in an ELFNOTE in the kernel binary.
502  *
503  * This function is passed to load_elf() when it is called from
504  * load_elfboot() which then additionally checks for an ELF Note of
505  * type XEN_ELFNOTE_PHYS32_ENTRY and passes it to this function to
506  * parse the PVH entry address from the ELF Note.
507  *
508  * Due to trickery in elf_opts.h, load_elf() is actually available as
509  * load_elf32() or load_elf64() and this routine needs to be able
510  * to deal with being called as 32 or 64 bit.
511  *
512  * The address of the PVH entry point is saved to the 'pvh_start_addr'
513  * global variable.  (although the entry point is 32-bit, the kernel
514  * binary can be either 32-bit or 64-bit).
515  */
516 static uint64_t read_pvh_start_addr(void *arg1, void *arg2, bool is64)
517 {
518     size_t *elf_note_data_addr;
519 
520     /* Check if ELF Note header passed in is valid */
521     if (arg1 == NULL) {
522         return 0;
523     }
524 
525     if (is64) {
526         struct elf64_note *nhdr64 = (struct elf64_note *)arg1;
527         uint64_t nhdr_size64 = sizeof(struct elf64_note);
528         uint64_t phdr_align = *(uint64_t *)arg2;
529         uint64_t nhdr_namesz = nhdr64->n_namesz;
530 
531         elf_note_data_addr =
532             ((void *)nhdr64) + nhdr_size64 +
533             QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
534 
535         pvh_start_addr = *elf_note_data_addr;
536     } else {
537         struct elf32_note *nhdr32 = (struct elf32_note *)arg1;
538         uint32_t nhdr_size32 = sizeof(struct elf32_note);
539         uint32_t phdr_align = *(uint32_t *)arg2;
540         uint32_t nhdr_namesz = nhdr32->n_namesz;
541 
542         elf_note_data_addr =
543             ((void *)nhdr32) + nhdr_size32 +
544             QEMU_ALIGN_UP(nhdr_namesz, phdr_align);
545 
546         pvh_start_addr = *(uint32_t *)elf_note_data_addr;
547     }
548 
549     return pvh_start_addr;
550 }
551 
552 static bool load_elfboot(const char *kernel_filename,
553                          int kernel_file_size,
554                          uint8_t *header,
555                          size_t pvh_xen_start_addr,
556                          FWCfgState *fw_cfg)
557 {
558     uint32_t flags = 0;
559     uint32_t mh_load_addr = 0;
560     uint32_t elf_kernel_size = 0;
561     uint64_t elf_entry;
562     uint64_t elf_low, elf_high;
563     int kernel_size;
564 
565     if (ldl_p(header) != 0x464c457f) {
566         return false; /* no elfboot */
567     }
568 
569     bool elf_is64 = header[EI_CLASS] == ELFCLASS64;
570     flags = elf_is64 ?
571         ((Elf64_Ehdr *)header)->e_flags : ((Elf32_Ehdr *)header)->e_flags;
572 
573     if (flags & 0x00010004) { /* LOAD_ELF_HEADER_HAS_ADDR */
574         error_report("elfboot unsupported flags = %x", flags);
575         exit(1);
576     }
577 
578     uint64_t elf_note_type = XEN_ELFNOTE_PHYS32_ENTRY;
579     kernel_size = load_elf(kernel_filename, read_pvh_start_addr,
580                            NULL, &elf_note_type, &elf_entry,
581                            &elf_low, &elf_high, NULL, 0, I386_ELF_MACHINE,
582                            0, 0);
583 
584     if (kernel_size < 0) {
585         error_report("Error while loading elf kernel");
586         exit(1);
587     }
588     mh_load_addr = elf_low;
589     elf_kernel_size = elf_high - elf_low;
590 
591     if (pvh_start_addr == 0) {
592         error_report("Error loading uncompressed kernel without PVH ELF Note");
593         exit(1);
594     }
595     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ENTRY, pvh_start_addr);
596     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, mh_load_addr);
597     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, elf_kernel_size);
598 
599     return true;
600 }
601 
602 void x86_load_linux(X86MachineState *x86ms,
603                     FWCfgState *fw_cfg,
604                     int acpi_data_size,
605                     bool pvh_enabled)
606 {
607     bool linuxboot_dma_enabled = X86_MACHINE_GET_CLASS(x86ms)->fwcfg_dma_enabled;
608     uint16_t protocol;
609     int setup_size, kernel_size, cmdline_size;
610     int dtb_size, setup_data_offset;
611     uint32_t initrd_max;
612     uint8_t header[8192], *setup, *kernel;
613     hwaddr real_addr, prot_addr, cmdline_addr, initrd_addr = 0;
614     FILE *f;
615     char *vmode;
616     MachineState *machine = MACHINE(x86ms);
617     struct setup_data *setup_data;
618     const char *kernel_filename = machine->kernel_filename;
619     const char *initrd_filename = machine->initrd_filename;
620     const char *dtb_filename = machine->dtb;
621     const char *kernel_cmdline = machine->kernel_cmdline;
622     SevKernelLoaderContext sev_load_ctx = {};
623 
624     /* Align to 16 bytes as a paranoia measure */
625     cmdline_size = (strlen(kernel_cmdline) + 16) & ~15;
626 
627     /* load the kernel header */
628     f = fopen(kernel_filename, "rb");
629     if (!f) {
630         fprintf(stderr, "qemu: could not open kernel file '%s': %s\n",
631                 kernel_filename, strerror(errno));
632         exit(1);
633     }
634 
635     kernel_size = get_file_size(f);
636     if (!kernel_size ||
637         fread(header, 1, MIN(ARRAY_SIZE(header), kernel_size), f) !=
638         MIN(ARRAY_SIZE(header), kernel_size)) {
639         fprintf(stderr, "qemu: could not load kernel '%s': %s\n",
640                 kernel_filename, strerror(errno));
641         exit(1);
642     }
643 
644     /* kernel protocol version */
645     if (ldl_p(header + 0x202) == 0x53726448) {
646         protocol = lduw_p(header + 0x206);
647     } else {
648         /*
649          * This could be a multiboot kernel. If it is, let's stop treating it
650          * like a Linux kernel.
651          * Note: some multiboot images could be in the ELF format (the same of
652          * PVH), so we try multiboot first since we check the multiboot magic
653          * header before to load it.
654          */
655         if (load_multiboot(x86ms, fw_cfg, f, kernel_filename, initrd_filename,
656                            kernel_cmdline, kernel_size, header)) {
657             return;
658         }
659         /*
660          * Check if the file is an uncompressed kernel file (ELF) and load it,
661          * saving the PVH entry point used by the x86/HVM direct boot ABI.
662          * If load_elfboot() is successful, populate the fw_cfg info.
663          */
664         if (pvh_enabled &&
665             load_elfboot(kernel_filename, kernel_size,
666                          header, pvh_start_addr, fw_cfg)) {
667             fclose(f);
668 
669             fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE,
670                 strlen(kernel_cmdline) + 1);
671             fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline);
672 
673             fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, sizeof(header));
674             fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA,
675                              header, sizeof(header));
676 
677             /* load initrd */
678             if (initrd_filename) {
679                 GMappedFile *mapped_file;
680                 gsize initrd_size;
681                 gchar *initrd_data;
682                 GError *gerr = NULL;
683 
684                 mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
685                 if (!mapped_file) {
686                     fprintf(stderr, "qemu: error reading initrd %s: %s\n",
687                             initrd_filename, gerr->message);
688                     exit(1);
689                 }
690                 x86ms->initrd_mapped_file = mapped_file;
691 
692                 initrd_data = g_mapped_file_get_contents(mapped_file);
693                 initrd_size = g_mapped_file_get_length(mapped_file);
694                 initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
695                 if (initrd_size >= initrd_max) {
696                     fprintf(stderr, "qemu: initrd is too large, cannot support."
697                             "(max: %"PRIu32", need %"PRId64")\n",
698                             initrd_max, (uint64_t)initrd_size);
699                     exit(1);
700                 }
701 
702                 initrd_addr = (initrd_max - initrd_size) & ~4095;
703 
704                 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
705                 fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
706                 fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data,
707                                  initrd_size);
708             }
709 
710             option_rom[nb_option_roms].bootindex = 0;
711             option_rom[nb_option_roms].name = "pvh.bin";
712             nb_option_roms++;
713 
714             return;
715         }
716         protocol = 0;
717     }
718 
719     if (protocol < 0x200 || !(header[0x211] & 0x01)) {
720         /* Low kernel */
721         real_addr    = 0x90000;
722         cmdline_addr = 0x9a000 - cmdline_size;
723         prot_addr    = 0x10000;
724     } else if (protocol < 0x202) {
725         /* High but ancient kernel */
726         real_addr    = 0x90000;
727         cmdline_addr = 0x9a000 - cmdline_size;
728         prot_addr    = 0x100000;
729     } else {
730         /* High and recent kernel */
731         real_addr    = 0x10000;
732         cmdline_addr = 0x20000;
733         prot_addr    = 0x100000;
734     }
735 
736     /* highest address for loading the initrd */
737     if (protocol >= 0x20c &&
738         lduw_p(header + 0x236) & XLF_CAN_BE_LOADED_ABOVE_4G) {
739         /*
740          * Linux has supported initrd up to 4 GB for a very long time (2007,
741          * long before XLF_CAN_BE_LOADED_ABOVE_4G which was added in 2013),
742          * though it only sets initrd_max to 2 GB to "work around bootloader
743          * bugs". Luckily, QEMU firmware(which does something like bootloader)
744          * has supported this.
745          *
746          * It's believed that if XLF_CAN_BE_LOADED_ABOVE_4G is set, initrd can
747          * be loaded into any address.
748          *
749          * In addition, initrd_max is uint32_t simply because QEMU doesn't
750          * support the 64-bit boot protocol (specifically the ext_ramdisk_image
751          * field).
752          *
753          * Therefore here just limit initrd_max to UINT32_MAX simply as well.
754          */
755         initrd_max = UINT32_MAX;
756     } else if (protocol >= 0x203) {
757         initrd_max = ldl_p(header + 0x22c);
758     } else {
759         initrd_max = 0x37ffffff;
760     }
761 
762     if (initrd_max >= x86ms->below_4g_mem_size - acpi_data_size) {
763         initrd_max = x86ms->below_4g_mem_size - acpi_data_size - 1;
764     }
765 
766     fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_ADDR, cmdline_addr);
767     fw_cfg_add_i32(fw_cfg, FW_CFG_CMDLINE_SIZE, strlen(kernel_cmdline) + 1);
768     fw_cfg_add_string(fw_cfg, FW_CFG_CMDLINE_DATA, kernel_cmdline);
769     sev_load_ctx.cmdline_data = (char *)kernel_cmdline;
770     sev_load_ctx.cmdline_size = strlen(kernel_cmdline) + 1;
771 
772     if (protocol >= 0x202) {
773         stl_p(header + 0x228, cmdline_addr);
774     } else {
775         stw_p(header + 0x20, 0xA33F);
776         stw_p(header + 0x22, cmdline_addr - real_addr);
777     }
778 
779     /* handle vga= parameter */
780     vmode = strstr(kernel_cmdline, "vga=");
781     if (vmode) {
782         unsigned int video_mode;
783         const char *end;
784         int ret;
785         /* skip "vga=" */
786         vmode += 4;
787         if (!strncmp(vmode, "normal", 6)) {
788             video_mode = 0xffff;
789         } else if (!strncmp(vmode, "ext", 3)) {
790             video_mode = 0xfffe;
791         } else if (!strncmp(vmode, "ask", 3)) {
792             video_mode = 0xfffd;
793         } else {
794             ret = qemu_strtoui(vmode, &end, 0, &video_mode);
795             if (ret != 0 || (*end && *end != ' ')) {
796                 fprintf(stderr, "qemu: invalid 'vga=' kernel parameter.\n");
797                 exit(1);
798             }
799         }
800         stw_p(header + 0x1fa, video_mode);
801     }
802 
803     /* loader type */
804     /*
805      * High nybble = B reserved for QEMU; low nybble is revision number.
806      * If this code is substantially changed, you may want to consider
807      * incrementing the revision.
808      */
809     if (protocol >= 0x200) {
810         header[0x210] = 0xB0;
811     }
812     /* heap */
813     if (protocol >= 0x201) {
814         header[0x211] |= 0x80; /* CAN_USE_HEAP */
815         stw_p(header + 0x224, cmdline_addr - real_addr - 0x200);
816     }
817 
818     /* load initrd */
819     if (initrd_filename) {
820         GMappedFile *mapped_file;
821         gsize initrd_size;
822         gchar *initrd_data;
823         GError *gerr = NULL;
824 
825         if (protocol < 0x200) {
826             fprintf(stderr, "qemu: linux kernel too old to load a ram disk\n");
827             exit(1);
828         }
829 
830         mapped_file = g_mapped_file_new(initrd_filename, false, &gerr);
831         if (!mapped_file) {
832             fprintf(stderr, "qemu: error reading initrd %s: %s\n",
833                     initrd_filename, gerr->message);
834             exit(1);
835         }
836         x86ms->initrd_mapped_file = mapped_file;
837 
838         initrd_data = g_mapped_file_get_contents(mapped_file);
839         initrd_size = g_mapped_file_get_length(mapped_file);
840         if (initrd_size >= initrd_max) {
841             fprintf(stderr, "qemu: initrd is too large, cannot support."
842                     "(max: %"PRIu32", need %"PRId64")\n",
843                     initrd_max, (uint64_t)initrd_size);
844             exit(1);
845         }
846 
847         initrd_addr = (initrd_max - initrd_size) & ~4095;
848 
849         fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_ADDR, initrd_addr);
850         fw_cfg_add_i32(fw_cfg, FW_CFG_INITRD_SIZE, initrd_size);
851         fw_cfg_add_bytes(fw_cfg, FW_CFG_INITRD_DATA, initrd_data, initrd_size);
852         sev_load_ctx.initrd_data = initrd_data;
853         sev_load_ctx.initrd_size = initrd_size;
854 
855         stl_p(header + 0x218, initrd_addr);
856         stl_p(header + 0x21c, initrd_size);
857     }
858 
859     /* load kernel and setup */
860     setup_size = header[0x1f1];
861     if (setup_size == 0) {
862         setup_size = 4;
863     }
864     setup_size = (setup_size + 1) * 512;
865     if (setup_size > kernel_size) {
866         fprintf(stderr, "qemu: invalid kernel header\n");
867         exit(1);
868     }
869     kernel_size -= setup_size;
870 
871     setup  = g_malloc(setup_size);
872     kernel = g_malloc(kernel_size);
873     fseek(f, 0, SEEK_SET);
874     if (fread(setup, 1, setup_size, f) != setup_size) {
875         fprintf(stderr, "fread() failed\n");
876         exit(1);
877     }
878     if (fread(kernel, 1, kernel_size, f) != kernel_size) {
879         fprintf(stderr, "fread() failed\n");
880         exit(1);
881     }
882     fclose(f);
883 
884     /* append dtb to kernel */
885     if (dtb_filename) {
886         if (protocol < 0x209) {
887             fprintf(stderr, "qemu: Linux kernel too old to load a dtb\n");
888             exit(1);
889         }
890 
891         dtb_size = get_image_size(dtb_filename);
892         if (dtb_size <= 0) {
893             fprintf(stderr, "qemu: error reading dtb %s: %s\n",
894                     dtb_filename, strerror(errno));
895             exit(1);
896         }
897 
898         setup_data_offset = QEMU_ALIGN_UP(kernel_size, 16);
899         kernel_size = setup_data_offset + sizeof(struct setup_data) + dtb_size;
900         kernel = g_realloc(kernel, kernel_size);
901 
902         stq_p(header + 0x250, prot_addr + setup_data_offset);
903 
904         setup_data = (struct setup_data *)(kernel + setup_data_offset);
905         setup_data->next = 0;
906         setup_data->type = cpu_to_le32(SETUP_DTB);
907         setup_data->len = cpu_to_le32(dtb_size);
908 
909         load_image_size(dtb_filename, setup_data->data, dtb_size);
910     }
911 
912     /*
913      * If we're starting an encrypted VM, it will be OVMF based, which uses the
914      * efi stub for booting and doesn't require any values to be placed in the
915      * kernel header.  We therefore don't update the header so the hash of the
916      * kernel on the other side of the fw_cfg interface matches the hash of the
917      * file the user passed in.
918      */
919     if (!sev_enabled()) {
920         memcpy(setup, header, MIN(sizeof(header), setup_size));
921     }
922 
923     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_ADDR, prot_addr);
924     fw_cfg_add_i32(fw_cfg, FW_CFG_KERNEL_SIZE, kernel_size);
925     fw_cfg_add_bytes(fw_cfg, FW_CFG_KERNEL_DATA, kernel, kernel_size);
926     sev_load_ctx.kernel_data = (char *)kernel;
927     sev_load_ctx.kernel_size = kernel_size;
928 
929     fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_ADDR, real_addr);
930     fw_cfg_add_i32(fw_cfg, FW_CFG_SETUP_SIZE, setup_size);
931     fw_cfg_add_bytes(fw_cfg, FW_CFG_SETUP_DATA, setup, setup_size);
932     sev_load_ctx.setup_data = (char *)setup;
933     sev_load_ctx.setup_size = setup_size;
934 
935     if (sev_enabled()) {
936         sev_add_kernel_loader_hashes(&sev_load_ctx, &error_fatal);
937     }
938 
939     option_rom[nb_option_roms].bootindex = 0;
940     option_rom[nb_option_roms].name = "linuxboot.bin";
941     if (linuxboot_dma_enabled && fw_cfg_dma_enabled(fw_cfg)) {
942         option_rom[nb_option_roms].name = "linuxboot_dma.bin";
943     }
944     nb_option_roms++;
945 }
946 
947 void x86_isa_bios_init(MemoryRegion *isa_bios, MemoryRegion *isa_memory,
948                        MemoryRegion *bios, bool read_only)
949 {
950     uint64_t bios_size = memory_region_size(bios);
951     uint64_t isa_bios_size = MIN(bios_size, 128 * KiB);
952 
953     memory_region_init_alias(isa_bios, NULL, "isa-bios", bios,
954                              bios_size - isa_bios_size, isa_bios_size);
955     memory_region_add_subregion_overlap(isa_memory, 1 * MiB - isa_bios_size,
956                                         isa_bios, 1);
957     memory_region_set_readonly(isa_bios, read_only);
958 }
959 
960 void x86_bios_rom_init(X86MachineState *x86ms, const char *default_firmware,
961                        MemoryRegion *rom_memory, bool isapc_ram_fw)
962 {
963     const char *bios_name;
964     char *filename;
965     int bios_size;
966     ssize_t ret;
967 
968     /* BIOS load */
969     bios_name = MACHINE(x86ms)->firmware ?: default_firmware;
970     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
971     if (filename) {
972         bios_size = get_image_size(filename);
973     } else {
974         bios_size = -1;
975     }
976     if (bios_size <= 0 ||
977         (bios_size % 65536) != 0) {
978         goto bios_error;
979     }
980     memory_region_init_ram(&x86ms->bios, NULL, "pc.bios", bios_size,
981                            &error_fatal);
982     if (sev_enabled()) {
983         /*
984          * The concept of a "reset" simply doesn't exist for
985          * confidential computing guests, we have to destroy and
986          * re-launch them instead.  So there is no need to register
987          * the firmware as rom to properly re-initialize on reset.
988          * Just go for a straight file load instead.
989          */
990         void *ptr = memory_region_get_ram_ptr(&x86ms->bios);
991         load_image_size(filename, ptr, bios_size);
992         x86_firmware_configure(ptr, bios_size);
993     } else {
994         memory_region_set_readonly(&x86ms->bios, !isapc_ram_fw);
995         ret = rom_add_file_fixed(bios_name, (uint32_t)(-bios_size), -1);
996         if (ret != 0) {
997             goto bios_error;
998         }
999     }
1000     g_free(filename);
1001 
1002     /* map the last 128KB of the BIOS in ISA space */
1003     x86_isa_bios_init(&x86ms->isa_bios, rom_memory, &x86ms->bios,
1004                       !isapc_ram_fw);
1005 
1006     /* map all the bios at the top of memory */
1007     memory_region_add_subregion(rom_memory,
1008                                 (uint32_t)(-bios_size),
1009                                 &x86ms->bios);
1010     return;
1011 
1012 bios_error:
1013     fprintf(stderr, "qemu: could not load PC BIOS '%s'\n", bios_name);
1014     exit(1);
1015 }
1016